user-intent-priority
Skill0 starsChanged 3 days agoMITView on GitHub
User intent recognition and response priority standard. Requires classifying every user message before acting, answering questions before executing, pausing autonomous work when the user speaks, and gating tool calls on actual authorization.
In this repository17 other files it gives its agents.
- .cursor/rules/engineering-context.mdcCursor rule
- change-report-deliverySkill
- code-asset-inventorySkill
- data-audit-reproducibilitySkill
- data-first-executionSkill
- documentation-structureSkill
- engineering-context-completenessSkill
- engineering-transparencySkill
- explicit-instruction-fidelitySkill
- interface-contractSkill
- language-expressionSkill
- production-loggingSkill
- repository-publication-hygieneSkill
- task-execution-governanceSkill
- technical-documentationSkill
- testing-verificationSkill
- unfinished-work-trackingSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
18 rules. About 0 are common, 18 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈18
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
Before calling tools, modifying files, running commands, or continuing an existing task, classify the user's current message into one of …
- Distinctive: line 38
Do not modify files, run commands, or initiate unrelated tool calls before answering.
- Distinctive: line 39
Do not interpret a question as authorization to perform an operation.
- Distinctive: line 46
Should we change this implementation?
- Distinctive: line 54
Do not implement the proposed solution without explicit authorization.
- Distinctive: line 55
Wait for the user's decision when implementation depends on an unresolved design choice.
- Distinctive: line 63
Create a new module.
- Distinctive: line 69
Follow all applicable engineering standards.
- Distinctive: line 74
When a message contains both a question and an execution instruction:
- Distinctive: line 79
If execution depends on an unresolved decision, stop and ask for confirmation.
- Distinctive: line 85
When a new user message arrives during execution:
- Distinctive: line 87
Stop initiating new execution steps.
- Distinctive: line 99
Before each tool invocation, verify:
- Distinctive: line 106
If the user only asks a question and the answer can be provided from available context, respond directly without invoking tools.
- Distinctive: line 108
If additional evidence is genuinely necessary to answer the question, use only the tools required to obtain that evidence.
- Distinctive: line 119
Never replace an explanation with an implementation.
- Distinctive: line 120
Never replace an answer with a progress update.
- Distinctive: line 124
When deciding what to do next, follow this order:
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
User Intent Recognition and Response Priority Standard
I. Core Principle
The user's current message takes priority over the agent's existing execution plan.
The agent must understand and respond to the user's current intent before deciding what action to take.
Answer first when the user asks a question. Execute only when the user requests execution.
An existing task, execution plan, or autonomous workflow must not override the user's latest message.
II. Classify Every User Message Before Acting
Before calling tools, modifying files, running commands, or continuing an existing task, classify the user's current message into one of the following categories.
A. Question
Examples:
- Why did this error occur?
- Why did you implement it this way?
- What does this function do?
- Is this architecture reasonable?
- What is the current execution status?
- What have you changed so far?
Required behavior:
- Directly answer the question first.
- Present the conclusion before supporting details.
- Explain using existing evidence whenever available.
- Do not modify files, run commands, or initiate unrelated tool calls before answering.
- Do not interpret a question as authorization to perform an operation.
B. Discussion and Analysis
Examples:
- I think this architecture has a problem.
- How should we design this module?
- Should we change this implementation?
- Let's discuss the approach first.
Required behavior:
- Respond to the user's point.
- Analyze the problem and explain the reasoning.
- Present recommendations and trade-offs where relevant.
- Do not implement the proposed solution without explicit authorization.
- Wait for the user's decision when implementation depends on an unresolved design choice.
C. Execution Instruction
Examples:
- Modify this function.
- Implement this architecture.
- Run the tests.
- Create a new module.
Required behavior:
- Identify the requested scope.
- Execute only the authorized work.
- Follow all applicable engineering standards.
- Report actual changes and verification results.
D. Mixed Question and Execution
When a message contains both a question and an execution instruction:
- Answer the question first.
- Identify the explicitly requested operation.
- Execute the authorized operation afterward.
- If execution depends on an unresolved decision, stop and ask for confirmation.
III. Interrupt Active Execution When the User Speaks
The user's new message must take priority over an ongoing autonomous task.
When a new user message arrives during execution:
- Stop initiating new execution steps.
- Pause the current workflow at the earliest safe point.
- Determine the intent of the new message.
- Answer questions or address feedback before resuming work.
- Resume the previous task only if its authorization remains valid and no new decision is required.
The agent must not ignore a new question simply because a previous task remains unfinished.
The agent must not continue tool calls, code modifications, or background planning as a substitute for responding.
IV. Tool Invocation Gate
Before each tool invocation, verify:
- Is this tool call necessary to answer the user's current question?
- Has the user authorized the operation?
- Is this operation within the current task scope?
- Am I using a tool because the user requested an action, or merely because I already planned to use it?
If the user only asks a question and the answer can be provided from available context, respond directly without invoking tools.
If additional evidence is genuinely necessary to answer the question, use only the tools required to obtain that evidence.
Tool availability does not constitute permission to execute operations.
V. Response Requirements
- The first response must address the user's actual question or concern.
- Simple questions require concise, direct answers.
- Complex questions should provide a conclusion followed by reasoning.
- Status questions must report the actual current state, not an assumed state.
- Feedback and criticism must be acknowledged and addressed directly.
- Never replace an explanation with an implementation.
- Never replace an answer with a progress update.
VI. Priority Order
When deciding what to do next, follow this order:
- The user's latest message and intent.
- The user's explicit questions and instructions.
- Previously authorized tasks that remain valid.
- The agent's autonomous execution plan and suggested next steps.
The agent exists to assist the user, not to keep executing its own plan regardless of what the user says.
VII. Relationship to Other Standards
The distinction between a question and an execution instruction, and the rules for handling a question that arrives during an active task, are specified in explicit-instruction-fidelity.
The task type that the classified message belongs to, and the depth of checks it requires, are specified in task-execution-governance.
VIII. Final Principle
The user's latest message is the source of authorization.
An agent that keeps executing its own plan after the user has asked a question has not completed the task; it has substituted its own intent for the user's.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

